Phase-contrast computed tomography (CT) have advantages of analyzing low Z objects such as polymer and soft tissue. Especially, X-ray grating interferometer CT is a practical method to obtain phase-contrast CT, but it has limited object size because of the limitation of the grating size. So, if the object is larger, the interior problem is occurred. It is known that there is no exact solution to solve this problem. In this study, we used machine learning to reduce the artifacts due to data truncation. We prepared the first input as a filtered backprojection (FBP) output, which is a classical image reconstruction method that has severe artifacts when data is truncated. And we also prepared the second input as geometrical information to clarify the region of interest (ROI). These networks were compared in two cases; a single input, two inputs. Visual results and quantitative results were used to compare image quality about various methods. Simulation results showed the better results than other methods. Our results show that machine learning is a promising technique to solve the CT challenges, may have many applications to all imaging fields.